Data-driven discharge analysis: a case study for the Wernersbach catchment, Germany
More details
Hide details
Institute of Physical Geography, Goethe University, Frankfurt am Main, Germany
Institute of Hydrology and Meteorology, Technische Universität Dresden, Dresden, Germany.
Corresponding author
Eklavyya Popat   

Institute of Physical Geography, Goethe University, Frankfurt am Main, Germany
Meteorology Hydrology and Water Management, 8(1),54–62
This study focuses on precipitation-discharge data-driven models, with regression analysis between the weighted maximum rainfall and maximum discharge of flood events. It is also the first of its kind investigation for the Wernersbach catchment, which incorporates data-driven models in order to evaluate the suitability of the model in simulating the discharge from the catchment and provide good insights for future studies. The input parameters are hydrological and climate data collected from 2001 to 2009, including precipitation, rainfall-runoff and soil moisture. The statistical regression and artificial neural network models used are based on a data-driven multiple linear regression technique, and the same input parameters are applied for validation and calibration. The artificial neural network model has one hidden layer with a sigmoidal activation function and uses a linear activation function in the output layer. The artificial neural network is observed to model 0.7% and 0.5% of values, with and without extreme values respectively. With less than 1% error, the artificial neural network is observed to predict extreme events better compared to the conventional statistical regression model and is also better suited to the tasks of rainfall-runoff and flood forecasting. It is presumed that in the future this study’s conclusions would form the basis for more complex and detailed studies for the same catchment area.